AI Automation / AI Agents
Create an agent that joins meetings (or reads transcripts), extracts action items, and distributes summaries.
Why This Prompt Exists
Meetings produce decisions and action items — but the value is lost when no one writes them down, or writes down the wrong things.
You get:
- action items that are too vague (“follow up” — with whom? by when?)
- missing decisions (what did we actually agree on?)
- no accountability (who owns each action?)
- summaries that are too long (no one reads them)
- summaries that are too short (missing critical context)
But meeting summaries can be structured:
- decisions: what was agreed upon?
- action items: who does what by when?
- topics discussed: what was covered?
- key questions: what’s still unresolved?
- speaker attribution: who said what (for clarity)
Without structure, meeting summaries are useless.
This prompt designs effective meeting summary agents.
The Prompt
Assume the role of a meeting productivity architect who designs AI summary agents.
Your task is to create an agent that extracts decisions and action items from meetings.
Generate:
1. MEETING INPUT
- Source: [Live transcript / recorded audio / meeting notes / chat log]
- Meeting type: [Standup / Planning / Retrospective / Client call / Internal review]
2. SUMMARY STRUCTURE
**Meeting Summary: [Title]**
- Date: [date]
- Attendees: [list]
- Duration: [actual vs. planned]
**Key Decisions**
- Decision 1: [what was decided] — Context: [why]
- Decision 2: ...
**Action Items**
| Action | Owner | Due Date | Priority |
|--------|-------|----------|----------|
| [task] | [person] | [date] | High/Med/Low |
**Topics Discussed** (brief)
- Topic 1: [key point]
- Topic 2: [key point]
**Unresolved Questions**
- Question 1: [what still needs an answer]
3. ACTION ITEM EXTRACTION RULES
- Must have clear owner (extract from "I'll do X" statements)
- Must have due date (if not stated, flag as "due date needed")
- Must be specific (not "work on" but "write proposal")
4. DECISION EXTRACTION RULES
- Look for consensus statements ("we agree that...", "sounds good", "let's do that")
- Capture dissenting opinions (if any)
- Note if decision was final or tentative
5. SPEAKER ATTRIBUTION (optional)
- For sensitive meetings, attribute key points to speakers
- Format: "[Name]: [point]"
6. OUTPUT DISTRIBUTION
- Where to send: [Slack channel, email, project management tool]
- Who approves: [meeting owner / none / automated]
7. PRIVACY RULES
- Never record or distribute: [sensitive topics, HR conversations]
- Redaction rules: [e.g., remove personal phone numbers, addresses]
8. READY-TO-USE AGENT PROMPT
- The system prompt for the meeting summary agent
INPUTS:
Meeting type and typical attendees:
[E.G., "Weekly engineering standup — 8 people"]
Meeting source:
[LIVE TRANSCRIPT / RECORDING / NOTES]
Confidentiality level:
[HIGH (legal/HR) / MEDIUM (internal) / LOW (public)]
Desired summary length:
[BRIEF (5 bullets) / STANDARD (1-2 pages) / DETAILED (full transcript with highlights)]
RULES:
- Action items without owners are not action items (flag them)
- Distinguish between decisions (what we agreed) and discussions (what we talked about)
- For recurring meetings, include a "carryover" section for incomplete actions
- Get consent before recording or transcribing meetings
- Comply with consent laws (one-party vs. two-party consent states)
- For sensitive meetings, have a human review before distribution
How To Use It
- Get consent before recording or transcribing meetings — comply with local laws.
- For sensitive meetings, have a human review the summary before distribution.
- Action items without owners or due dates are useless — flag them for follow-up.
- Distinguish between decisions (what was agreed) and discussions (what was talked about).
- For recurring meetings, include a “carryover” section for incomplete actions.
Example Input
Meeting type:
“Weekly product planning — 6 people (PM, dev leads, design)”
Meeting source:
“Live transcript from Zoom”
Confidentiality level:
“MEDIUM”
Desired summary length:
“STANDARD”
Why It Works
Most meeting summaries are either verbatim transcripts (too long) or vague notes (too short) — missing the critical action items and decisions.
This framework improves outcomes by forcing:
- decisions extraction (what was agreed?)
- action item specification (who does what by when?)
- topic summary (what was covered?)
- unresolved questions (what still needs answers?)
- distribution rules (who gets the summary?)
Great meeting summary agents don’t just transcribe — they transform conversation into action.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.